Learning Vision-Based Flight in Drone Swarms by Imitation

Abstract

Decentralized drone swarms deployed today either rely on sharing of positions among agents or detecting swarm members with the help of visual markers. This work proposes an entirely visual approach to coordinate markerless drone swarms based on imitation learning. Each agent is controlled by a small and efficient convolutional neural network that takes raw omnidirectional images as inputs and predicts 3D velocity commands that match those computed by a flocking algorithm. We start training in simulation and propose a simple yet effective unsupervised domain adaptation approach to transfer the learned controller to the real world. We further train the controller with data collected in our motion capture hall. We show that the convolutional neural network trained on the visual inputs of the drone can learn not only robust inter-agent collision avoidance but also cohesion of the swarm in a sample-efficient manner. The neural controller effectively learns to localize other agents in the visual input, which we show by visualizing the regions with the most influence on the motion of an agent. We remove the dependence on sharing positions among swarm members by taking only local visual information into account for control. Our work can therefore be seen as the first step towards a fully decentralized, vision-based swarm without the need for communication or visual markers.

A model of collective behavior based purely on vision

“…Recently, the visual projection field has appeared as a central feature of collective movements in fish (11)(12)(13)(14), birds (15), humans (16), or artificial systems (17,18). Because of the geometrical nature of vision, i.e., the projection of the environment, vision appears as a good starting point to explore the relationship between sensory information and emergent collective behaviors.…”

Predictive Control of Aerial Swarms in Cluttered Environments

“…In robotics, the advantages of the NMPC method are promising for applications that require navigation in crowded scenarios, such as the exploration of urban environments, collapsed buildings, or forests [45], [46]. Also vision-based swarms could benefit from all these features since the reliability of reciprocal visual detection of the drones strongly depends on their distance, and NMPC swarms showed that they can better maintain target inter-agent distances [22], [47]. Overall, predictive methods can improve the autonomy of swarm operations as well as the safety of the swarm and the environment, which are both essential elements to build public confidence in the use of swarms [48].…”

“…These rules consist of (a) cohesion, which brings each agent closer to its neighbors, (b) repulsion, which drives each agent away from its neighbors to avoid collisions, and (c) alignment, which steers each agent towards the average heading of its neighbors. In goal-directed flight, alignment is replaced by migration, which steers each agent in a preferred migration direction [21], [22]. For navigating environments with obstacles, the addition of a fourth rule, collision avoidance, is necessary to steer the agents around the obstacles [20], [23], [24].…”

The Influence of Limited Visual Sensing on the Reynolds Flocking Algorithm

“…For example, a decentralized and distributed approach can make the system easily scalable and robust to failures. For this reason, the implementation of swarms based on these principles has received a lot of attention in recent years [21], [22].…”